A building 3D printing material proportioning self-adaptive optimization preparation method

Abstract

The application discloses a kind of building 3D printing material proportioning self-adaptive optimization preparation methods, belong to intelligent construction and building additive manufacturing technical field.This method is aimed at the industry pain points such as poor generalization of existing proportioning design, insufficient open-loop control working condition adaptation, difficult coordination of multi-objective performance, poor solid waste adaptability, constructs "data acquisition-intelligent optimization-simulation verification-preparation execution-feedback correction-model iteration" full closed-loop system, embeds material physical constitutive equation into artificial intelligence model, completes proportioning self-adaptive optimization by the depth reinforcement learning of physical constraint, realizes dynamic correction in combination with digital twin preplay and full-process closed-loop feedback.This application can realize stable and controllable material performance under complex conditions, greatly improve the solid waste content and printing component qualification rate, adapt to building 3D printing industrialization and large-scale application.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction and building additive manufacturing technology, specifically to a method for adaptive optimization of the formulation of building 3D printing materials. It is applicable to various mainstream building 3D printing material systems such as cement-based, geopolymer-based, and solid waste composite-based materials, and can be widely used in various building 3D printing engineering scenarios such as prefabricated building component printing, irregular building structure printing, municipal facility printing, and green solid waste building material printing. It is especially suitable for industrial building 3D printing projects with high requirements for material performance stability, printing accuracy, preparation efficiency, and low carbon and environmental protection. Background Technology

[0002] As a core supporting technology for new building industrialization and intelligent construction, 3D printing technology has expanded from printing non-load-bearing components to large-scale engineering applications such as load-bearing walls and entire buildings, thanks to its significant advantages such as moldless molding, high design freedom, high material utilization, and short construction cycle. 3D printing materials, as the core carrier of this technology, directly determine the printability, mechanical properties, durability, and molding accuracy of the materials through their rational formulation, which in turn determines the safety, stability, and economy of the printed structure.

[0003] Current technologies for the formulation and preparation of 3D printing materials for construction still face several core bottlenecks that restrict large-scale application in the industry. These bottlenecks are as follows:

[0004] First, the proportioning design lacks underlying physical mechanism support, resulting in insufficient generalization ability and engineering feasibility. Existing proportioning optimization methods mostly adopt empirical trial-and-error methods or pure data-driven black-box machine learning models, without embedding the basic physical laws of material hydration and rheological evolution. The proportioning schemes output by the models often violate the basic laws of materials science. Although the laboratory fitting effect is excellent, the generalization ability is extremely poor under the complex working conditions of the construction site, and it cannot be directly applied.

[0005] Second, the open-loop control is the primary method and cannot cope with real-time changes in dynamic working conditions. Existing technologies mostly adopt an open-loop process of "laboratory trial mixing - field application," which only conducts raw material testing once before preparation. This cannot respond to dynamic disturbances such as fluctuations in aggregate moisture content, changes in ambient temperature and humidity, differences in batch performance of admixtures, and adjustments to printing process parameters at the construction site. Fixed proportions are prone to defects such as pipe blockage, mold collapse, interlayer delamination, and cracking, resulting in a large amount of material waste and component scrap.

[0006] Third, the focus is mainly on single-point performance control, which cannot achieve multi-objective synergistic optimization. Existing formulation optimizations mostly focus on improving a single performance, or prioritize material mechanical strength at the expense of printing accuracy, or prioritize flowability optimization while ignoring stacking stability. They cannot take into account the performance requirements of the entire process of pumping, extrusion, stacking, and hardening, and it is difficult to achieve multi-objective synergistic optimization of printability, mechanical properties, molding accuracy, and material cost.

[0007] Fourth, the existing proportioning methods are poorly adapted to industrial solid waste and cannot meet the requirements of green and low-carbon development. The existing proportioning methods have not designed specific optimization logic for the material characteristics of industrial solid wastes such as fly ash, mineral powder, steel slag, and recycled aggregates. The solid waste content is generally less than 20%, and high content can easily lead to a significant decline in material performance, which cannot give full play to the value of solid waste resource utilization and does not meet the "dual carbon" development goals of the construction industry.

[0008] Fifth, the implementation of intelligent and engineering-based systems is insufficient, and the implementation threshold is high. Existing intelligent proportioning systems mostly rely on dedicated testing equipment and high-performance computing platforms, which result in high equipment modification costs and high operating thresholds, making them difficult for small and medium-sized enterprises to apply. At the same time, there is a lack of standardized preparation execution processes, and the proportioning scheme is disconnected from the actual preparation process. The theoretical proportions deviate significantly from the actual output performance, making it impossible to achieve large-scale and continuous engineering applications.

[0009] While existing technologies have made some attempts to improve upon the aforementioned issues, none have formed a systematic solution. Some solutions use a single algorithm for proportion prediction, failing to address the core deficiency of black-box models lacking physical constraints; others only achieve online adjustment of a single parameter, unable to achieve global optimization under multi-variable coupling; still others have complex technical architectures with extremely high implementation barriers, unable to adapt to the complex environment and engineering requirements of building construction sites. Therefore, developing an adaptive optimization preparation method for the proportion of building 3D printing materials, which combines underlying mechanism support, closed-loop management throughout the entire process, multi-objective collaborative optimization, high solid waste adaptability, and strong engineering feasibility, has become a pressing technical challenge in this field. Summary of the Invention

[0010] The purpose of this invention is to overcome the aforementioned deficiencies of existing technologies and provide a method for adaptive optimization of material proportions in 3D printing of buildings. Based on the constitutive laws of material physics, this method constructs a full-process, closed-loop adaptive optimization system for proportions, fundamentally addressing the core pain points of existing technologies, such as poor generalization of black-box models, insufficient adaptability to open-loop control conditions, difficulty in coordinating multi-objective performance, poor adaptability to solid waste, and weak engineering feasibility. The core objective of this invention is to achieve real-time adaptive matching of the proportioning scheme to dynamic changes in raw materials, environment, and processes, ensuring that the material remains stably within the optimal printable range under complex conditions, while simultaneously considering mechanical properties, molding accuracy, and green, low-carbon attributes. This provides a complete technical solution for the large-scale, high-quality, and industrial application of 3D printing technology in buildings.

[0011] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: a method for adaptively optimizing the formulation of building 3D printing materials, comprising the following steps:

[0012] S1 offline basic database and physical constraint digital twin foundation construction;

[0013] S2 Real-time Acquisition and Standardized Preprocessing of Multi-Source Heterogeneous Data;

[0014] S3 Physical Constraint Embedding Deep Reinforcement Learning Matching Adaptive Optimization;

[0015] S4 Multiphysics Coupled Digital Twin Pre-simulation and Matching Feasibility Verification;

[0016] S5 full-process closed-loop feedback and dynamic ratio correction;

[0017] S6 adaptive proportioning precision execution and automated preparation;

[0018] S7 model incremental self-learning and iterative database updates.

[0019] Further, step S1 specifically includes:

[0020] S11 conducts cross-experiments in material rheology, hydration kinetics, and mechanical properties for the target building 3D printing material system, and constructs a dataset of mapping relationships between proportion, rheological parameters, printing performance, and mechanical properties; the material system includes at least one of cement-based, geopolymer-based, and solid waste composite-based materials.

[0021] Based on the classical constitutive equations of rheology, the differential equations of cement hydration kinetics, and the thermal-humid-mechanical coupling equations of interlayer interfaces, S12 constructs a set of physical constraint equations for the evolution of material properties. In the multidimensional rheological parameter space, it defines a rheological safety domain that simultaneously meets the requirements of the entire process of pumping, extrusion, stacking, and hardening. The rheological safety margin is defined by the shortest Mahalanobis distance from the current state point to the boundary of the safety domain, thus quantifying the robustness of the material's printability.

[0022] S13 Based on the dataset and physical constraint equations, a digital twin engine embedded with physical laws is constructed. The engine integrates a discrete element-computational fluid dynamics (DEM-CFD) coupled solver and an early strength finite element prediction module, which can simulate the extrusion molding process, stacking stability and mechanical property development law of the proportioning scheme under the target working conditions.

[0023] S14 constructs a standardized parameter library, including a raw material performance threshold library, an environmental condition boundary library, a printing process parameter library, and a component performance standard library, forming an offline basic database to provide benchmark support for online proportioning optimization.

[0024] Furthermore, in step S11, the detection parameters of the cross-test include: dynamic yield stress, plastic viscosity, static yield stress, thixotropic recovery rate, setting time, slump, spread, 7d / 28d compressive strength, interlaminar bond strength, and drying shrinkage; the boundary conditions of the rheological safety zone are: dynamic yield stress 80~250Pa, thixotropic recovery time 3~15s, initial setting time 3~8h, and 28d compressive strength ≥ 1.1 times the target design strength.

[0025] Further, step S2 specifically includes:

[0026] The S21 uses a multimodal sensor array to collect four categories of dynamic parameters in real time: ① basic properties of raw materials, ② environmental conditions, ③ printing equipment operating parameters, and ④ design parameters of the target component.

[0027] S22 performs millisecond-level hardware timestamp alignment on all collected data, uses the 3σ criterion combined with sliding window filtering to remove outliers, eliminates random interference through wavelet denoising algorithm, and then performs min-max normalization to output standardized feature vectors.

[0028] S23 performs multi-scale reconstruction of time-series features, constructs a hybrid feature tensor containing instantaneous values, statistical features, and changing trends, and inputs it into the matching optimization model.

[0029] Furthermore, in step S21:

[0030] Basic properties of raw materials include cementitious material activity index, specific surface area, setting time, aggregate particle size distribution, mud content, moisture content, admixture solid content, effective ingredient ratio, and industrial solid waste admixture activity index and particle size distribution; sampling frequency is no less than once per batch, and the detection accuracy of key parameters is ≤±1%;

[0031] Environmental operating parameters include ambient temperature, relative humidity, wind speed, and solar radiation intensity, with monitoring accuracies of ±0.5℃, ±2%RH, ±0.1m / s, and ±5W / ㎡, respectively, and a sampling frequency of no less than 1 time / minute;

[0032] The operating parameters of the printing equipment include extrusion pressure, extrusion rate, printing speed, printing layer thickness, nozzle diameter, path curvature, and stirring speed, with monitoring accuracies of ±0.05MPa, ±0.1L / min, ±0.1mm / s, and ±0.05mm, respectively, and a sampling frequency of not less than 10Hz;

[0033] The design parameters of the target component include design strength grade, interlayer bonding requirements, surface roughness tolerance, and geometric complexity index, which are automatically extracted through BIM model analysis.

[0034] Furthermore, step S3 specifically includes:

[0035] S31 constructs a hybrid optimization model with physical constraint embedding. The model consists of a multi-scale feature encoding network, a Physical Information Neural Network (PINN) constraint layer, and a Deep Deterministic Policy Gradient (DDPG) reinforcement learning module connected in series.

[0036] The multi-scale feature coding network described in S32 receives standardized feature vectors, extracts nonlinear coupled features with different dimensional parameters through a multi-branch convolutional network, and outputs a hidden state representation.

[0037] The PINN constraint layer described in S33 incorporates a set of physical constraint equations for material hydration dynamics, rheological constitutives, and interface coupling. The residuals of the equation set serve as the core component of the loss function, constraining the feature space and ensuring that the output sizing candidate set conforms to the basic laws of materials science and engineering boundary conditions.

[0038] The DDPG reinforcement learning module described in S34 operates with an Actor-Critic architecture. The Actor network outputs the ratio action vector, and the Critic network evaluates the long-term cumulative reward of the current ratio under the corresponding state. With the goal of maximizing rheological safety margin, minimizing molding quality deviation, maximizing mechanical performance compliance rate, and minimizing material cost as multi-objective optimization objectives, a reward function is constructed to complete global optimization within the feasible region constrained by PINN and output the initial optimal ratio scheme.

[0039] Further, in step S33, the loss function expression for the PINN constraint layer is: in, To account for the fitting error of the proportion-performance data, The sum of squared residuals of the physical constraint equations. Penalty items for violations of engineering boundary conditions, This represents the initial material state matching error. The weights are adaptive and decrease dynamically with each training round.

[0040] Furthermore, in step S34:

[0041] The proportioning action vectors include water-cement ratio, sand-ash ratio, water-reducing agent dosage, viscosity modifier dosage, accelerator / retarder dosage, industrial solid waste admixture dosage, and fiber volume dosage.

[0042] The reward function expression is:

[0043] in, Rheological safety margin score, Scoring is given based on whether the mechanical properties meet the standards. To score the molding accuracy, Scoring for material costs, Penalties for violating engineering boundary conditions; These are weighting coefficients, summing to 1, and can be adaptively adjusted according to the printing scenario.

[0044] Furthermore, step S4 specifically includes:

[0045] S41 inputs the initial optimal ratio scheme into the digital twin engine, substitutes the real-time collected environmental conditions and printing process parameters, and carries out multi-physics field coupling simulation to simulate the material's performance evolution throughout the entire process from stirring, pumping, extrusion to deposition and hardening.

[0046] S42 uses the boundary of the rheological safety domain and the performance requirements of component design as verification thresholds to determine the feasibility of the simulation results: if the simulation results meet all threshold requirements, a feasible mix design that passes verification is output; if the simulation results exceed the threshold range, a constraint violation flag is triggered, a negative reward is fed back to the reinforcement learning module, and the mix design optimization is carried out again until a feasible mix design that meets the verification requirements is output.

[0047] Further, step S5 specifically includes:

[0048] In the entire process of material preparation and printing construction, S51 collects process data in real time through an online monitoring array, including material rheological parameters during the stirring process, pipeline pressure changes during the pumping process, strip cross-sectional morphology during the extrusion process, and component forming quality and surface temperature rise rate after printing.

[0049] S52 performs residual analysis on the measured process data and the digital twin simulation results to calculate the comprehensive deviation coefficient. When the deviation coefficient exceeds the preset threshold, the online correction subroutine is started: the state variables of the PINN constraint layer are updated by the Extended Kalman Filter (EKF), the online fine-tuning mode of the DDPG module is activated, the proportion correction amount is solved within the engineering boundary constraints, and the updated proportion scheme is generated.

[0050] The single correction of the S53 formulation scheme shall not exceed 8% of the initial formulation, and the consecutive corrections shall not exceed 3 times to avoid system oscillation; the corrected formulation scheme shall be simultaneously sent to the preparation execution system to achieve dynamic closed-loop control of the entire process.

[0051] Further, step S6 specifically includes:

[0052] The S61 automated batching and preparation system receives instructions on the batching scheme and uses high-precision metering equipment to accurately measure each component of the raw materials, with the metering accuracy controlled within ±0.2%.

[0053] S62 adopts a three-stage standardized mixing process:

[0054] ① In the dry material premixing stage, cementitious materials, aggregates, solid waste admixtures, and dry powder additives are put into a twin-shaft forced mixer and dry-mixed at a speed of 20~40 rpm for 60~90 seconds.

[0055] ② In the liquid phase mixing stage, pump in the mixing water and liquid additives according to the ratio, and switch to a speed of 40~70 rpm for wet mixing for 120~180 seconds;

[0056] ③ During the rheological stabilization stage, the rotation speed is reduced to 10~20 rpm and the material is allowed to stand for 30~60 seconds. During this period, the rheological parameters of the material are monitored in real time through the built-in micro rheological probe. If the material deviates from the target range by ±10%, the micro-compensation injection valve is triggered for dynamic fine-tuning.

[0057] The material, after being mixed by the S63, is fed into a high-pressure screw pump via a screw conveyor. It is then transported to the 3D printing equipment hopper while maintaining a constant back pressure. The conveying pressure and speed are adaptively adjusted according to the material's flowability to prevent material segregation and blockage.

[0058] Further, step S7 specifically includes:

[0059] After each printing batch is completed, the S71 system automatically archives the entire process data, including raw material parameters, environmental conditions, mixing scheme, process monitoring data, and actual performance data of components, forming a standardized sample.

[0060] S72 When the cumulative number of newly added samples reaches 50 batches, or the comprehensive deviation coefficient of consecutive batches exceeds 5%, the model iterative update is triggered: the transfer learning strategy is used to fine-tune the hybrid optimization model, and the material constitutive parameter library and offline basic database of the digital twin engine are updated simultaneously.

[0061] The S73 model update uses the elastic average stochastic gradient descent algorithm to ensure consistency between cloud training and edge deployment, enabling the model to learn from slow time-varying factors such as material batch changes and equipment wear and tear.

[0062] Furthermore, the building 3D printing material systems compatible with this method include:

[0063] (1) Cement-based printing materials: Ordinary Portland cement is used as the main cementing material. The optimization focuses on the water-cement ratio, sand ratio, and admixture dosage. The water-cement ratio is controlled at 0.30~0.45, and the sand ratio is controlled at 35%~45%.

[0064] (2) Geopolymer-based printing materials: slag and fly ash are the main cementing materials, and alkali activator is the core additive. The optimization focuses on the modulus and dosage of alkali activator, the proportion of cementing materials, and the water-cement ratio. The concentration of alkali activator is controlled at 30%~50%, and the mass ratio of slag to fly ash is controlled at 1:1~3:1.

[0065] (3) Solid waste composite printing material: fly ash, mineral powder, steel slag powder and recycled aggregate are the main raw materials. The total amount of industrial solid waste is controlled at 30%~50%. The optimization focus is on the matching of solid waste content, the gradation of recycled aggregate and the compatibility of additives, so as to ensure that the mechanical properties and printability of the material meet the standards in a coordinated manner.

[0066] Compared with the prior art, the present invention has the following outstanding substantive features and significant progress:

[0067] 1. Underlying mechanism support, completely solving the engineering feasibility pain points of black box models.

[0068] This invention innovatively embeds material physics constitutive equations into an artificial intelligence model. Through a PINN constraint layer, the model output is forced to conform to the fundamental laws of hydration kinetics and rheology. This fundamentally avoids the shortcomings of purely data-driven models, which often exhibit excellent fit in the laboratory but are unusable in the field. The model's generalization ability and engineering feasibility are significantly improved. Experimental verification shows that the formulation schemes output by this invention achieve a 100% first-time pass rate in the field, requiring no secondary manual adjustments.

[0069] 2. Closed-loop management throughout the entire process enables real-time adaptive matching under complex operating conditions.

[0070] This invention constructs a fully closed-loop control system of "data acquisition - intelligent optimization - simulation verification - preparation execution - feedback correction - model iteration", which can respond in real time to dynamic disturbances such as raw material fluctuations, changes in environmental temperature and humidity, and adjustments to printing processes. The ratio optimization response time is ≤15s and the optimization accuracy is ≥98%, which completely solves the problem of poor adaptability under fixed ratio working conditions. It can stably control the material performance deviation within ±2% and improve the qualified rate of printed components to over 99%.

[0071] 3. Multi-objective global collaborative optimization, taking into account the performance requirements of the entire process.

[0072] This invention, with rheological safety margin as its core, constructs a multi-objective optimized reward function that considers printability, mechanical properties, molding accuracy, and material cost. It achieves synergistic performance across the entire process of pumping, extrusion, deposition, and hardening, overcoming the shortcomings of existing technologies that prioritize one aspect over another. Experimental verification shows that the material prepared using this invention exhibits a 28-day compressive strength margin ≥1.1, interlayer bond strength ≥3.5 MPa, dimensional error of printed components ≤2.0%, and drying shrinkage ≤0.02%, simultaneously meeting the dual requirements of high strength and high precision.

[0073] 4. High adaptability to solid waste, meeting the needs of green and low-carbon development.

[0074] This invention designs a dedicated optimization logic for the material characteristics of industrial solid waste. It incorporates solid waste dosage adaptation constraints during the proportion optimization process, enabling stable application of industrial solid waste with a total dosage of 30% to 50%, which is far higher than the dosage level of less than 20% in existing technologies. This significantly reduces the amount of cement and natural sand and gravel used, thereby reducing carbon emissions. At the same time, through multi-component synergistic optimization, it avoids the decline in material performance under high solid waste dosage, achieving a win-win situation of environmental protection and practicality.

[0075] 5. Strong engineering feasibility, suitable for large-scale industrial applications.

[0076] This invention adopts a standardized and modular system architecture. All hardware consists of existing mature equipment in the construction industry, eliminating the need for specialized equipment development. Existing 3D printing systems can be applied simply by upgrading the software and adding a few sensors, resulting in low modification costs and low implementation barriers. It is also equipped with a standardized three-stage preparation process with a measurement accuracy of ≤±0.2%. The entire preparation process is automated without human intervention, reducing material waste to less than 3%. The preparation efficiency is more than 50% higher than existing technologies, making it directly adaptable to industrialized and continuous 3D printing engineering applications in construction.

[0077] 6. Lifelong self-learning capability, continuous evolution of system performance.

[0078] This invention establishes an incremental model iteration mechanism that can continuously optimize model parameters and digital twin engine based on actual engineering data. The model prediction accuracy and adaptability continuously improve with the duration of use, effectively cope with slow time-varying factors such as material batch changes and equipment wear and tear, and has the ability to operate stably for a long time. This provides sustainable technical support for the long-term large-scale application of building 3D printing technology. Attached Figure Description

[0079] Figure 1 This is a flowchart illustrating the overall process framework of the adaptive optimization preparation method for building 3D printing materials according to the present invention.

[0080] Figure 2 This is a diagram showing the internal architecture and data flow of the hybrid optimization model with embedded physical constraints as described in this invention.

[0081] Figure 3 This is a schematic diagram of the hardware deployment of the multi-source data acquisition and full-process closed-loop feedback control system described in this invention;

[0082] Figure 4 This is a three-stage standardized process flow diagram of the automated preparation execution system described in this invention. Detailed Implementation

[0083] The present invention will be further described in detail below with reference to specific embodiments. These embodiments are only for illustrating the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Equivalent substitutions or partial adjustments to parameters, equipment, and processes made by those skilled in the art without departing from the core concept of the present invention should all fall within the scope of protection of the present invention.

[0084] The raw materials, equipment, and testing methods used in this embodiment are all commercially available products and industry standard methods in the field, and comply with the relevant technical specifications for 3D printing of buildings.

[0085] Example 1: Adaptive optimization of printing material ratio for C40 cement-based load-bearing wall components

[0086] This embodiment uses the printing of load-bearing walls in high-rise residential buildings as an application scenario. The target material is C40 cement-based building 3D printing material, and the preparation method of this invention is applied. The specific steps are as follows:

[0087] Step 1: Construction of Offline Base Database and Physically Constrained Digital Twin Underlying

[0088] For the P.O42.5 ordinary silicate cement-based material system, 120 sets of cross-tests were conducted to construct a dataset of proportion-rheology-performance mapping relationships. The experimental variables included water-cement ratio (0.30~0.45), sand ratio (35%~45%), polycarboxylate superplasticizer dosage (0.8%~1.5%), and fly ash dosage (10%~30%). Based on the Herschel-Bulkley rheological constitutive equation and the cement hydration kinetic differential equation, a set of physical constraint equations was constructed to define the rheological safety domain boundary of C40 material: dynamic yield stress 100~220Pa, thixotropic recovery time 4~12s, initial setting time 4~6h, and 28d compressive strength ≥44MPa. A digital twin engine embedded with physical constraints was constructed, along with a C40 load-bearing component performance standard library, and offline substrate construction was completed.

[0089] Step 2: Real-time acquisition and standardized preprocessing of multi-source heterogeneous data

[0090] Collect parameters across all dimensions using a multimodal sensor array:

[0091] ① Raw material parameters: P.O42.5 cement specific surface area 365m² / kg, 28-day activity index 93%; river sand particle size 0.15~5mm, mud content 2.1%, moisture content 3.2%; Grade II fly ash specific surface area 320m² / kg, 28-day activity index 82%; polycarboxylate superplasticizer solid content 30%, water reduction rate 28%;

[0092] ②Environmental conditions: The ambient temperature at the construction site was 22℃, the relative humidity was 62%, the wind speed was 1.0m / s, and the sampling frequency was 1 time / minute;

[0093] ③ Equipment operating parameters: Extrusion pressure of the printing equipment is 2.8MPa, printing speed is 60mm / s, printing layer thickness is 5mm, nozzle diameter is 20mm, and path curvature is 0~0.5m. -1 The sampling frequency is 10Hz;

[0094] ④ Component design parameters: Design strength grade C40, interlayer bond strength ≥4.0MPa, surface roughness tolerance ≤0.5mm / m, geometric complexity index 0.2.

[0095] All data undergoes hardware timestamp alignment, outlier removal, wavelet denoising, and normalization to output standardized feature vectors, which are then input into the hybrid optimization model.

[0096] Step 3: Adaptive Optimization of Deep Reinforcement Learning Allocation for Physical Constraint Embedding

[0097] After receiving the feature vector, the hybrid optimization model extracts coupled features through a multi-scale feature encoding network, performs physical feasibility filtering through a PINN constraint layer, and then completes global optimization through the DDPG reinforcement learning module. The weight coefficients for this scenario are set as follows: =0.35 (rheological safety) =0.3 (mechanical properties) =0.2 (molding precision) =0.1 (cost) =0.05 (penalty). The final output initial optimal mix design (mass fraction) is: cement 24%, fly ash 12%, river sand 41%, crushed stone 18%, polycarboxylate superplasticizer 1.2%, and water 3.8%.

[0098] Step 4: Multiphysics Coupled Digital Twin Pre-simulation and Matching Feasibility Verification

[0099] The initial mix design was input into the digital twin engine, and DEM-CFD coupled simulation and strength prediction were performed by substituting real-time working parameters. The simulation results showed that the material dynamic yield stress was 150 Pa, thixotropic recovery time was 6 s, initial setting time was 4.8 h, and 28-day compressive strength was 46.2 MPa, all of which met the rheological safety domain and design performance requirements. The verification was successful, and a feasible mix design was output.

[0100] Step 5: Closed-loop feedback and dynamic adjustment of proportions throughout the entire process

[0101] During the preparation and printing process, data was collected in real time through an online monitoring array: the material slump was 220 mm and the expansion was 360 mm during stirring; the pipeline pressure was stable at 2.8 MPa during pumping; and the cross-sectional morphology of the extruded strip deviated from the design value by ≤1.5%. Comparing the measured data with the twin simulation results, the overall deviation coefficient was 1.2%, which is lower than the preset threshold of 2.0%. No ratio correction was required, and the current ratio scheme was maintained for continued preparation.

[0102] Simulating extreme conditions: The ambient temperature suddenly dropped to 8℃, and the initial setting time of the material was observed to extend to 7.2 hours, exceeding the boundary of the rheological safety domain, with a comprehensive deviation coefficient reaching 6.8%, triggering an online correction subroutine. The PINN state variables were updated via EKF, and the DDPG module output corrective proportions: accelerator dosage increased to 0.6%, water-reducing agent dosage decreased to 1.0%, and the water-cement ratio was fine-tuned to 0.36. After correction, a re-simulation verification showed that the initial setting time of the material recovered to 5.2 hours, meeting the safety domain requirements. The correction was then sent to the preparation system for execution, achieving dynamic closed-loop control.

[0103] Step 6: Adaptive Proportioning Precision Execution and Automated Preparation

[0104] The automated preparation system receives the proportioning instructions and executes them in batches of 1000 kg.

[0105] ① High-precision metering: 240kg cement, 120kg fly ash, 410kg river sand, 180kg crushed stone, 12kg polycarboxylate superplasticizer, 38kg water, metering accuracy ≤ ±0.2%;

[0106] ② Three-stage mixing: dry material premixing stage: 30 rpm for 75 seconds; liquid phase mixing stage: 60 rpm for 150 seconds; rheological stabilization stage: 15 rpm for 45 seconds. During this period, the built-in rheological probe monitors in real time, ensuring stable performance without deviation.

[0107] ③ Pumping and conveying: The material is conveyed to the printing equipment hopper at a speed of 3m / min and a back pressure of 2.5MPa. The conveying process is stable and there is no segregation.

[0108] Step 7: Incremental self-learning of the model and iterative updates of the database

[0109] A total of 8 batches were printed, producing a total of 8000 kg. After printing, the wall components were tested, and the results showed: 28-day compressive strength of 45.8 MPa, interlayer bond strength of 4.3 MPa, average dimensional deviation of 1.8%, and no cracking, mold collapse, or interlayer delamination defects, all meeting design requirements. The entire process data was archived to the basic database. After accumulating 50 new samples, model fine-tuning was triggered, reducing the model prediction error from 1.8% to below 1.5%, completing the iterative update.

[0110] Example 2: Adaptive optimization of the formulation of geopolymer-based irregular decorative component printing materials

[0111] This embodiment uses the printing of irregularly shaped architectural decorative components as the application scenario. The target material is a geopolymer-based architectural 3D printing material, and the preparation method of this invention is applied. The specific steps are as follows:

[0112] Step 1: Construction of Offline Base Database and Physically Constrained Digital Twin Underlying

[0113] Ninety cross-experiments were conducted on slag-fly ash based polymer material systems to construct a dataset of proportion-rheology-performance mapping relationships. The experimental variables included slag / fly ash mass ratio (1:1~3:1), alkali activator modulus (1.2~1.8), alkali equivalent (8%~14%), and water-cement ratio (0.30~0.40). Based on the rheological constitutive equation and the hydration kinetic equation of alkali-activated cementitious materials, a set of physical constraint equations was constructed, defining the rheological safety domain boundary as follows: dynamic yield stress 80~200 Pa, thixotropic recovery time 3~10 s, initial setting time 3~5 h, and 28-day compressive strength ≥35 MPa. A digital twin engine and a performance standard library for irregularly shaped components were constructed, and offline substrate construction was completed.

[0114] Step 2: Real-time acquisition and standardized preprocessing of multi-source heterogeneous data

[0115] Collect core parameters:

[0116] ① Raw material parameters: S95 slag specific surface area 400m² / kg, 28d activity index 95%; Grade I fly ash specific surface area 350m² / kg, 28d activity index 85%; sodium hydroxide-water glass composite alkali activator modulus 1.5, total concentration 40%; manufactured sand particle size 0.15~4.75mm, mud content 1.8%;

[0117] ②Environmental conditions: Indoor constant temperature 25℃, relative humidity 55%, wind speed 0.3m / s;

[0118] ③ Equipment operating parameters: Extrusion pressure 2.2MPa, printing speed 40mm / s, printing layer thickness 2mm, nozzle diameter 15mm, path curvature 0.2~2.0m -1 ;

[0119] ④ Component design parameters: design strength grade C35, surface roughness tolerance ≤0.3mm / m, dimensional error ≤1.5%, geometric complexity index 1.2.

[0120] After preprocessing, the data is input into the hybrid optimization model, with the weighting coefficients set as follows: =0.3、 =0.25、 =0.35、 =0.05、 =0.05, with a focus on ensuring molding accuracy. The final output initial optimal mix proportion (mass fraction) is: 30% slag, 15% fly ash, 38% manufactured sand, 12% composite alkali activator, and 5% water.

[0121] Step 3: Digital Twin Pre-visualization and Feasibility Verification

[0122] Simulation results show that the material's dynamic yield stress is 120 Pa, thixotropic recovery time is 5 s, initial setting time is 3.8 h, 28-day compressive strength is 38.5 MPa, and molding size deviation is ≤1.2%, all of which meet the verification requirements, and a feasible mix design is output.

[0123] Step 4: Closed-loop feedback and preparation execution

[0124] The preparation process employed a three-stage mixing technique: dry mixing at 35 rpm for 60 seconds, wet mixing at 65 rpm for 120 seconds, and maturation at 15 rpm followed by standing for 30 seconds. The resulting material exhibited stable rheological properties. During printing, machine vision detected a 1.6% dimensional deviation in the irregular curved surface area, slightly exceeding the threshold. This triggered online correction, increasing the viscosity modifier dosage by 0.08% and fine-tuning the printing speed to 35 mm / s. After correction, the dimensional deviation returned to 1.3%, meeting the design requirements.

[0125] The final printed irregular-shaped component has a 28-day compressive strength of 39.2 MPa, an interlayer bond strength of 4.2 MPa, a surface roughness of 0.28 mm / m, and no obvious molding defects, fully meeting the requirements for the use of irregular-shaped decorative components.

[0126] Example 3: Adaptive optimization of the formulation of printing material for municipal sidewalk slabs with high solid waste content

[0127] This embodiment uses municipal sidewalk slab printing as an application scenario, with the target material being a high-content solid waste composite-based 3D printing material for buildings. The core objective is to achieve high-value utilization of industrial solid waste. The preparation method of this invention is applied, and the specific steps are as follows:

[0128] Step 1: Construction of Offline Base Database and Physically Constrained Digital Twin Underlying

[0129] For the fly ash-mineral powder-steel slag powder-recycled aggregate solid waste composite system, 100 sets of cross-experiments were conducted to construct a dataset of proportion-rheology-performance mapping relationships, with the core variables being the total solid waste content (30%~50%) and the recycled aggregate content (30%~60%). A set of physical constraint equations was constructed to define the boundary of the rheological safety domain: dynamic yield stress 90~230Pa, thixotropic recovery time 5~15s, initial setting time 5~7h, and 28d compressive strength ≥25MPa. A digital twin engine and a municipal component performance standard library were constructed, and offline substrate construction was completed.

[0130] Step 2: Real-time acquisition and standardized preprocessing of multi-source heterogeneous data

[0131] Collect core parameters:

[0132] ①Raw material parameters: P.O42.5 cement is used as an auxiliary cementitious material; the total content of fly ash, mineral powder and steel slag powder is 40%, and the 28-day activity indexes are 81%, 84% and 78% respectively; recycled aggregate accounts for 50% of the total aggregate, with fine recycled aggregate particle size of 0.15~5mm, coarse recycled aggregate particle size of 5~15mm, and mud content of 2.5%;

[0133] ② Environmental conditions: Outdoor ambient temperature 20℃, relative humidity 65%, wind speed 1.5m / s;

[0134] ③ Equipment operating parameters: extrusion pressure 3.0MPa, printing speed 70mm / s, printing layer thickness 4mm, nozzle diameter 25mm;

[0135] ④ Component design parameters: Design strength grade C25, 28d compressive strength ≥25MPa, dimensional error ≤3.0%, and wear resistance meets municipal sidewalk slab standards.

[0136] After preprocessing, the data is input into the hybrid optimization model, with the weighting coefficients set as follows: =0.3、 =0.3、 =0.15、 =0.2、 =0.05, focusing on ensuring the balance between solid waste content and mechanical properties. The final output initial optimal mix design (mass fraction) is: cement 18%, fly ash 14%, mineral powder 14%, steel slag powder 12% (total solid waste content 40%), recycled aggregate 25%, natural river sand 13%, water-retaining agent 1.0%, early strength agent 0.7%, and water 2.3%.

[0137] Step 3: Digital Twin Pre-visualization and Feasibility Verification

[0138] Simulation results show that the material's dynamic yield stress is 160 Pa, thixotropic recovery time is 8 s, initial setting time is 6.2 h, and 28-day compressive strength is 28.5 MPa, all of which meet the verification requirements, and a feasible mix design is output.

[0139] Step 4: Closed-loop feedback and preparation execution

[0140] The preparation process employs a three-stage stirring process, with a single batch production of 1000 kg and a metering accuracy of ≤±0.2%. The slump of the stirred material is 210 mm, the spread is 340 mm, and the rheological properties are stable. During the printing process, fluctuations in the moisture content of the recycled aggregate were detected, leading to a slight increase in material fluidity. This triggered a minor correction, reducing the water content by 0.2% and increasing the water-retaining agent content by 0.05%. After the correction, the material properties returned to the target range.

[0141] The final printed sidewalk slabs have a 28-day compressive strength of 29.1 MPa, an abrasion loss of ≤2.0 kg / m², a dimensional deviation of 2.2%, and no cracks or sandblasting defects. They fully meet the requirements for use in municipal facilities and also achieve a high content of 40% industrial solid waste, which significantly reduces material costs and carbon emissions.

[0142] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention, and are not actually limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for adaptively optimizing the formulation of building 3D printing materials, characterized in that: Includes the following steps: S1 constructs a dataset mapping the ratio, rheological parameters, printing performance, and mechanical properties of target building 3D printing material systems. Based on the physical constitutive laws of material rheology, hydration dynamics, and interface coupling, it constructs a set of physical constraint equations, defines the rheological safety domain and rheological safety margin for the entire process of material pumping, extrusion, stacking, and hardening, and constructs a digital twin engine and standardized parameter library embedded with physical laws, forming an offline basic database and a physical constraint digital twin base. S2 uses a multimodal sensor array to collect multi-source heterogeneous dynamic parameters in real time, including basic properties of raw materials, environmental conditions, printing equipment operation, and target component design. It performs time-series alignment, outlier removal, noise reduction, and normalization preprocessing on the collected data to construct a hybrid feature tensor containing instantaneous values, statistical characteristics, and trends. S3 constructs a hybrid optimization model with embedded physical constraints. The model consists of a multi-scale feature encoding network, a physical information neural network PINN constraint layer, and a deep deterministic policy gradient DDPG reinforcement learning module connected in series. The preprocessed hybrid feature tensor is input into the hybrid optimization model. After feature extraction and physical constraint filtering, the model completes multi-objective global optimization with the maximization of rheological safety margin as the core, and outputs the initial optimal ratio scheme. S4 inputs the initial optimal mix design into the digital twin engine, substitutes real-time operating parameters to carry out multi-physics coupling full-process simulation, and uses the rheological safety domain boundary and component design performance requirements as verification thresholds to determine feasibility. If the verification passes, a feasible mix design is output; if the verification fails, a negative reward is fed back and the mix optimization is carried out again. In the entire process of material preparation and printing, S5 collects process monitoring data in real time, performs residual analysis on the measured data and digital twin pre-simulation results and calculates the comprehensive deviation coefficient. When the deviation coefficient exceeds the preset threshold, it starts the online correction subroutine to solve the ratio correction amount, generates the updated ratio scheme and sends it out for execution, realizing dynamic closed-loop control of the entire process. S6, based on a feasible proportioning scheme, uses high-precision metering equipment to accurately measure the raw materials of each component, adopts a standardized stirring process to complete material preparation, and adaptively adjusts the conveying parameters according to the material flowability to achieve stable material delivery; after a single batch of printing is completed, S7 archives the entire process data to form a standardized sample, and when the preset trigger conditions are met, it adopts a transfer learning strategy to complete the incremental fine-tuning of the model, and simultaneously updates the offline basic database and digital twin engine to achieve lifelong self-learning of the model.

2. The method for adaptive optimization of the formulation of building 3D printing materials according to claim 1, characterized in that: In step S1, the boundary conditions of the rheological safety domain are: dynamic yield stress 80~250Pa, thixotropic recovery time 3~15s, initial setting time 3~8h, and 28d compressive strength ≥ 1.1 times the target design strength; the detection parameters of the mapping relationship dataset include dynamic yield stress, plastic viscosity, static yield stress, thixotropic recovery rate, setting time, slump, spread, 7d / 28d compressive strength, interlaminar bond strength, and drying shrinkage.

3. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S2, the basic attribute parameters of raw materials include the activity index of cementitious materials, the moisture content and gradation of aggregates, the proportion of effective components of admixtures, and the activity index of industrial solid waste admixtures; the sampling frequency of environmental operating condition parameters is not less than once per minute, the sampling frequency of printing equipment operating parameters is not less than 10Hz, and the design parameters of the target component are automatically extracted through BIM model analysis.

4. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S3, the PINN constraint layer incorporates a set of physical constraint equations for material hydration dynamics, rheological constitutive model, and interface coupling. Its loss function expression is as follows: in, To account for the fitting error of the proportion-performance data, The sum of squared residuals of the physical constraint equations. Penalty items for violations of engineering boundary conditions, This represents the initial material state matching error. These are adaptive weighting coefficients.

5. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S3, the ratio action vector output by the DDPG reinforcement learning module includes water-cement ratio, sand-ash ratio, admixture dosage, industrial solid waste admixture dosage, and fiber volume dosage. Its reward function expression is: in, Rheological safety margin score, Scoring is given based on whether the mechanical properties meet the standards. To score the molding accuracy, Scoring for material costs, Penalties for violating engineering boundary conditions; These are the weighting coefficients, and their sum is 1.

6. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S5, the single correction magnitude of the proportioning scheme shall not exceed 8% of the initial proportion, and the continuous correction shall not exceed 3 times; the online correction subroutine updates the state variables of the PINN constraint layer through extended Kalman filtering and activates the online fine-tuning mode of the DDPG module to solve for the proportioning correction amount.

7. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S6, the raw material metering accuracy of the high-precision metering equipment is controlled within ±0.2%.

8. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S6, the standardized mixing process is a three-stage process, specifically including: ① dry material premixing stage, dry mixing at 20~40 rpm for 60~90s; ② liquid phase mixing stage, wet mixing at 40~70 rpm for 120~180s; ③ rheological stabilization stage, static maturation at 10~20 rpm for 30~60s, during which material rheological parameters are monitored in real time and dynamic fine-tuning is triggered.

9. The adaptive optimization method for preparing building 3D printing materials according to claim 1, characterized in that: In step S7, the triggering condition for model iteration update is: the cumulative number of newly added samples reaches 50 batches, or the comprehensive deviation coefficient of consecutive batches exceeds 5%; the model update adopts the elastic average stochastic gradient descent algorithm.

10. The method for adaptive optimization of the formulation of building 3D printing materials according to claim 1, characterized in that: The building 3D printing material systems adapted to the method include cement-based, geopolymer-based, and solid waste composite-based materials. The total amount of industrial solid waste incorporated into the solid waste composite base material is controlled at 30% to 50%.

Images